Analysis by AI assistant
How much the assistants agree with each other
We compare which restaurants each assistant cites, how often, and whether they agree with each other. The reliable comparison is between DeepSeek and OpenAI with web search: OpenAI without search barely has usable data in this cut, explained below.
Summary
Of the restaurants mentioned by more than one assistant, only 2 are cited by all three, and 12 by exactly two. The vast majority, 484 restaurants, are cited by only one assistant. DeepSeek and OpenAI with web search have different favorite lists: Restaurante Mena, El Pegolí, and La Seu lead in DeepSeek; Quique Dacosta, Peix & Brases, and MESA La Marina lead in OpenAI with search.
The three assistants, in numbers
| Assistant | Calls | With capturable text | Mentions | Distinct restaurants | Mentions / usable call |
|---|---|---|---|---|---|
| DeepSeek (API) | 306 | 177 | 1052 | 401 | 5.94 |
| OpenAI, without web search | 88 | 2 | 7 | 7 | 3.5 |
| OpenAI, with web search | 57 | 57 | 219 | 106 | 3.84 |
How much the recommendations overlap
Of the 498 restaurants mentioned in total, only 2 appear on all three analyzed assistants' lists, 12 on exactly two, and 484 on just one. This suggests each assistant has its own set of go-to names for Dénia, rather than a shared consensus.
Why OpenAI without search isn't in the main comparison
Of the 88 calls to OpenAI without the web search tool, only 2 returned capturable text (see the data quality notice): the other 86 are part of the capture gap documented in the time-slot analysis. With 2 usable calls, no conclusion can be drawn about that assistant's real behavior in this cut.
What we can't claim from this
- Low overlap between assistants doesn't say which one is "right": each may draw on different training sources and search behaviors we can't observe from outside.
- DeepSeek accounts for 68% of this cut's calls (306 of 451): this isn't an equal-N comparison, so absolute mention counts aren't directly comparable, only the per-usable-call averages.
- This comparison doesn't include any provider's consumer interface, only its API: see GEO UI vs API for that distinction.
Keep exploring the dataset
This is one of six dimensions we analyze. The rest of the variables, and the full dataset to explore on your own, are on the data page.